Final Report Summary - CONVEXVISION (Convex Optimization Methods for Computer Vision and Image Analysis)
In the ERC project ConvexVision, we have developed convex optimization methods for a number of central challenges in computer vision. In particular, we have developed convex formulations for multiple view reconstruction, image segmentation, stereo reconstruction and optical flow, and semantic scene understanding. Furthermore we have developed novel primal-dual algorithms for efficiently minimizing convex and nonconvex optimization problems. As a result we can compute optimal or near-optimal solutions to classical problems where previous techniques only found locally optimal solutions.